Issues with CP-VASP Calculations for +U Systems

[springlau]

Dear all,

I hope this message finds you well. I am reaching out to discuss some issues I’ve encountered while performing calculations with CP-VASP version 1, particularly for systems that require a +U correction, such as surface catalysis on the Ce-based surface (with +U set to X).

We have successfully applied the +U values in our group and have published several papers, so I am confident that these values are reliable. Standard VASP calculations do not show any issues; however, when using CP-VASP, I am experiencing convergence problems, and the results appear highly anomalous. For instance, the calculated forces are exceedingly large, reaching values like 960.91873400, and the final energy calculations diverge significantly from those obtained with VASP.

To provide some context:

After single-point energy optimization, I have observed a force of 960.91873400 with a final energy of approximately -66339.416546. For comparison, a typical VASP calculation yields forces around 0.04672800 and energies around -806.512320. When we remove the +U correction and perform standard single-point energy calculations with CP-VASP, we still obtain similarly exaggerated results for both forces and total energy.

I have also tried adjusting the target potential values and the electron count multiple times, but these changes have not improved the situation.

Could you please help me understand what might be causing these discrepancies? Any insights you could provide regarding potential issues with the +U implementation in CP-VASP would be greatly appreciated.

Thank you for your assistance!

Best regards, Zhaochun

[ZLevell]

Dear Zhaochun

I suspect the anomalous energy and forces are due to the SCF cycle repeatedly failing to converge. Could you share your POSCAR, OUTCAR, and screen output with me? Could you also try doing the same fixed structure calculation with VASPsol turned on and CP-VASP turned off? I'm curious to see if the SCF will converge. If not, then it might make sense to relax the system without CP-VASP and to use its CHGCAR as an input for the CP-VASP simulation.

[springlau]

Dear ZLevell,

Thank you for your response. Attached are the files you requested. I suspect the anomalous energy and forces are indeed due to the SCF cycle repeatedly failing to converge. I have previously used VASPsol for single-point energy optimization without CP-VASP, and the structure energy and force were similar. I can't recall whether it converged or not, but the results were still significantly off.

I can take a look at pure vaspsol file again, but it may take me some time since it has been a while . You can check these documents first to see if there are any suggestions and comments.

[springlau]

The attachment is the result of pure vaspsol single point energy. I also used vasp to optimize the structure first, and vaspsol calculated the solvation single point energy. Although the number of atoms loaded is different, the Ce-based surface is completely consistent. . I don't find the same structure. But I have tested the calculated structures for many time. It is consistent. . The energy and force are abnormally large. . You can check whether it converges. . I thought SCF should converge. .

In addition, vaspsol cannot optimize the structure of this system and relax it. It will cause the structure to collapse directly or fly around. I have tried it and checked it.

[springlau]

Dear ZLevell, I really hope it can be solved as soon as possible, because the shared structure，Or if you have downloaded all, tell me and I will delete the structure！ thanks a lot!

[ZLevell]

I've downloaded your files. I will let you know once I've figured it out.

[springlau]

Thank you very much, Zachary. I have already deleted the files. I am not in a hurry anymore. I will wait patiently for your reply and important comments. 😄

[springlau]

Hi Zachary， is there any update for this issues?

[ZLevell]

Dear Zhaochun,

I apologize for the delayed response. Have you reached out to the VASPsol development team? Since the SCF cycle cannot be converged with regular VASPsol, it's unlikely that CP-VASP will be able to do so.

From the OSZICAR, it seems that the energy between each SCF step is changing rapidly and doesn't slow down. I think your best bet is to try another fixed structure calculation with regular VASPsol (no constant potential) with more accurate settings, such as ALGO = All and PREC = Accurate. Once you've gotten the SCF successfully converged, you can save the CHGCAR and use that as the initial input for CP-VASP, which should help it converge at the start.

I noticed ISMEAR is misspelled as "ISMAEAR", which is causing VASP to use the default value of 1. I also noticed that in the CP-VASP calculation, NELECT is set to 705, whereas the charge neutral value is 802. I'm unfamiliar with your system, but it should be easiest to start from the charge neutral value and then allow CP-VASP to tune the electron number. Did you get 705 from previous CP-VASP calculations? I suspect that may be due to incomplete SCF cycles and is an incorrect prediction.